]>
git.proxmox.com Git - mirror_lxc.git/blob - src/lxc/utils.c
2 * lxc: linux Container library
4 * (C) Copyright IBM Corp. 2007, 2008
7 * Daniel Lezcano <daniel.lezcano at free.fr>
9 * This library is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * This library is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with this library; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 #define __STDC_FORMAT_MACROS /* Required for PRIu64 to work. */
41 #include <sys/mount.h>
42 #include <sys/param.h>
43 #include <sys/prctl.h>
45 #include <sys/types.h>
50 #include "namespace.h"
55 #include "include/strlcpy.h"
59 #include "include/strlcat.h"
63 #define O_PATH 010000000
67 #define O_NOFOLLOW 00400000
70 lxc_log_define(utils
, lxc
);
73 * if path is btrfs, tries to remove it and any subvolumes beneath it
75 extern bool btrfs_try_remove_subvol(const char *path
);
77 static int _recursive_rmdir(const char *dirname
, dev_t pdev
,
78 const char *exclude
, int level
, bool onedev
)
80 struct dirent
*direntp
;
83 char pathname
[MAXPATHLEN
];
84 bool hadexclude
= false;
86 dir
= opendir(dirname
);
88 ERROR("failed to open %s", dirname
);
92 while ((direntp
= readdir(dir
))) {
96 if (!strcmp(direntp
->d_name
, ".") ||
97 !strcmp(direntp
->d_name
, ".."))
100 rc
= snprintf(pathname
, MAXPATHLEN
, "%s/%s", dirname
, direntp
->d_name
);
101 if (rc
< 0 || rc
>= MAXPATHLEN
) {
102 ERROR("pathname too long");
107 if (!level
&& exclude
&& !strcmp(direntp
->d_name
, exclude
)) {
108 ret
= rmdir(pathname
);
112 INFO("Not deleting snapshot %s", pathname
);
116 ret
= unlink(pathname
);
118 INFO("Failed to remove %s", pathname
);
121 SYSERROR("Failed to rmdir %s", pathname
);
129 ret
= lstat(pathname
, &mystat
);
131 ERROR("Failed to stat %s", pathname
);
136 if (onedev
&& mystat
.st_dev
!= pdev
) {
137 /* TODO should we be checking /proc/self/mountinfo for
138 * pathname and not doing this if found? */
139 if (btrfs_try_remove_subvol(pathname
))
140 INFO("Removed btrfs subvolume at %s\n", pathname
);
144 if (S_ISDIR(mystat
.st_mode
)) {
145 if (_recursive_rmdir(pathname
, pdev
, exclude
, level
+1, onedev
) < 0)
148 if (unlink(pathname
) < 0) {
149 SYSERROR("Failed to delete %s", pathname
);
155 if (rmdir(dirname
) < 0 && !btrfs_try_remove_subvol(dirname
) && !hadexclude
) {
156 ERROR("Failed to delete %s", dirname
);
162 ERROR("Failed to close directory %s", dirname
);
166 return failed
? -1 : 0;
169 /* In overlayfs, st_dev is unreliable. So on overlayfs we don't do the
172 static bool is_native_overlayfs(const char *path
)
174 if (has_fs_type(path
, OVERLAY_SUPER_MAGIC
) ||
175 has_fs_type(path
, OVERLAYFS_SUPER_MAGIC
))
181 /* returns 0 on success, -1 if there were any failures */
182 extern int lxc_rmdir_onedev(const char *path
, const char *exclude
)
187 if (is_native_overlayfs(path
))
190 if (lstat(path
, &mystat
) < 0) {
194 ERROR("Failed to stat %s", path
);
198 return _recursive_rmdir(path
, mystat
.st_dev
, exclude
, 0, onedev
);
201 /* borrowed from iproute2 */
202 extern int get_u16(unsigned short *val
, const char *arg
, int base
)
211 res
= strtoul(arg
, &ptr
, base
);
212 if (!ptr
|| ptr
== arg
|| *ptr
|| res
> 0xFFFF || errno
!= 0)
220 extern int mkdir_p(const char *dir
, mode_t mode
)
222 const char *tmp
= dir
;
223 const char *orig
= dir
;
227 dir
= tmp
+ strspn(tmp
, "/");
228 tmp
= dir
+ strcspn(dir
, "/");
230 makeme
= strndup(orig
, dir
- orig
);
232 if (mkdir(makeme
, mode
) && errno
!= EEXIST
) {
233 SYSERROR("failed to create directory '%s'", makeme
);
250 if (stat(RUNTIME_PATH
, &sb
) < 0)
253 if (geteuid() == sb
.st_uid
|| getegid() == sb
.st_gid
) {
254 rundir
= strdup(RUNTIME_PATH
);
258 rundir
= getenv("XDG_RUNTIME_DIR");
260 rundir
= strdup(rundir
);
264 INFO("XDG_RUNTIME_DIR isn't set in the environment.");
265 homedir
= getenv("HOME");
267 ERROR("HOME isn't set in the environment.");
271 rundir
= malloc(sizeof(char) * (17 + strlen(homedir
)));
275 sprintf(rundir
, "%s/.cache/lxc/run/", homedir
);
280 int wait_for_pid(pid_t pid
)
285 ret
= waitpid(pid
, &status
, 0);
296 if (!WIFEXITED(status
) || WEXITSTATUS(status
) != 0)
302 int lxc_wait_for_pid_status(pid_t pid
)
307 ret
= waitpid(pid
, &status
, 0);
322 #include <gnutls/gnutls.h>
323 #include <gnutls/crypto.h>
325 __attribute__((constructor
))
326 static void gnutls_lxc_init(void)
328 gnutls_global_init();
331 int sha1sum_file(char *fnam
, unsigned char *digest
)
341 f
= fopen_cloexec(fnam
, "r");
343 SYSERROR("Error opening template");
347 if (fseek(f
, 0, SEEK_END
) < 0) {
348 SYSERROR("Error seeking to end of template");
353 if ((flen
= ftell(f
)) < 0) {
354 SYSERROR("Error telling size of template");
359 if (fseek(f
, 0, SEEK_SET
) < 0) {
360 SYSERROR("Error seeking to start of template");
365 if ((buf
= malloc(flen
+1)) == NULL
) {
366 SYSERROR("Out of memory");
371 if (fread(buf
, 1, flen
, f
) != flen
) {
372 SYSERROR("Failure reading template");
379 SYSERROR("Failre closing template");
385 ret
= gnutls_hash_fast(GNUTLS_DIG_SHA1
, buf
, flen
, (void *)digest
);
391 struct lxc_popen_FILE
*lxc_popen(const char *command
)
396 struct lxc_popen_FILE
*fp
= NULL
;
398 ret
= pipe2(pipe_fds
, O_CLOEXEC
);
411 /* duplicate stdout */
412 if (pipe_fds
[1] != STDOUT_FILENO
)
413 ret
= dup2(pipe_fds
[1], STDOUT_FILENO
);
415 ret
= fcntl(pipe_fds
[1], F_SETFD
, 0);
421 /* duplicate stderr */
422 if (pipe_fds
[1] != STDERR_FILENO
)
423 ret
= dup2(pipe_fds
[1], STDERR_FILENO
);
425 ret
= fcntl(pipe_fds
[1], F_SETFD
, 0);
430 /* unblock all signals */
431 ret
= sigfillset(&mask
);
435 ret
= pthread_sigmask(SIG_UNBLOCK
, &mask
, NULL
);
439 execl("/bin/sh", "sh", "-c", command
, (char *)NULL
);
446 fp
= malloc(sizeof(*fp
));
450 memset(fp
, 0, sizeof(*fp
));
452 fp
->child_pid
= child_pid
;
453 fp
->pipe
= pipe_fds
[0];
455 /* From now on, closing fp->f will also close fp->pipe. So only ever
456 * call fclose(fp->f).
458 fp
->f
= fdopen(pipe_fds
[0], "r");
465 /* We can only close pipe_fds[0] if fdopen() didn't succeed or wasn't
466 * called yet. Otherwise the fd belongs to the file opened by fdopen()
467 * since it isn't dup()ed.
469 if (fp
&& !fp
->f
&& pipe_fds
[0] >= 0)
472 if (pipe_fds
[1] >= 0)
484 int lxc_pclose(struct lxc_popen_FILE
*fp
)
493 wait_pid
= waitpid(fp
->child_pid
, &wstatus
, 0);
494 } while (wait_pid
< 0 && errno
== EINTR
);
505 int randseed(bool srand_it
)
508 srand pre-seed function based on /dev/urandom
510 unsigned int seed
= time(NULL
) + getpid();
513 f
= fopen("/dev/urandom", "r");
515 int ret
= fread(&seed
, sizeof(seed
), 1, f
);
517 SYSDEBUG("unable to fread /dev/urandom, fallback to time+pid rand seed");
528 uid_t
get_ns_uid(uid_t orig
)
532 uid_t nsid
, hostid
, range
;
533 FILE *f
= fopen("/proc/self/uid_map", "r");
537 while (getline(&line
, &sz
, f
) != -1) {
538 if (sscanf(line
, "%u %u %u", &nsid
, &hostid
, &range
) != 3)
541 if (hostid
<= orig
&& hostid
+ range
> orig
) {
542 nsid
+= orig
- hostid
;
555 bool dir_exists(const char *path
)
560 ret
= stat(path
, &sb
);
562 /* Could be something other than eexist, just say "no". */
565 return S_ISDIR(sb
.st_mode
);
568 /* Note we don't use SHA-1 here as we don't want to depend on HAVE_GNUTLS.
569 * FNV has good anti collision properties and we're not worried
570 * about pre-image resistance or one-way-ness, we're just trying to make
571 * the name unique in the 108 bytes of space we have.
573 uint64_t fnv_64a_buf(void *buf
, size_t len
, uint64_t hval
)
577 for(bp
= buf
; bp
< (unsigned char *)buf
+ len
; bp
++)
579 /* xor the bottom with the current octet */
580 hval
^= (uint64_t)*bp
;
583 * multiply by the 64 bit FNV magic prime mod 2^64
585 hval
+= (hval
<< 1) + (hval
<< 4) + (hval
<< 5) +
586 (hval
<< 7) + (hval
<< 8) + (hval
<< 40);
592 bool is_shared_mountpoint(const char *path
)
594 char buf
[LXC_LINELEN
];
599 f
= fopen("/proc/self/mountinfo", "r");
603 while (fgets(buf
, LXC_LINELEN
, f
)) {
604 for (p
= buf
, i
= 0; p
&& i
< 4; i
++)
605 p
= strchr(p
+ 1, ' ');
609 p2
= strchr(p
+ 1, ' ');
614 if (strcmp(p
+ 1, path
) == 0) {
615 /* This is the path. Is it shared? */
616 p
= strchr(p2
+ 1, ' ');
617 if (p
&& strstr(p
, "shared:")) {
629 * Detect whether / is mounted MS_SHARED. The only way I know of to
630 * check that is through /proc/self/mountinfo.
631 * I'm only checking for /. If the container rootfs or mount location
632 * is MS_SHARED, but not '/', then you're out of luck - figuring that
633 * out would be too much work to be worth it.
635 int detect_shared_rootfs(void)
637 if (is_shared_mountpoint("/"))
642 bool switch_to_ns(pid_t pid
, const char *ns
)
645 char nspath
[MAXPATHLEN
];
647 /* Switch to new ns */
648 ret
= snprintf(nspath
, MAXPATHLEN
, "/proc/%d/ns/%s", pid
, ns
);
649 if (ret
< 0 || ret
>= MAXPATHLEN
)
652 fd
= open(nspath
, O_RDONLY
);
654 SYSERROR("Failed to open %s", nspath
);
660 SYSERROR("Failed to set process %d to %s of %d.", pid
, ns
, fd
);
670 * looking at fs/proc_namespace.c, it appears we can
671 * actually expect the rootfs entry to very specifically contain
672 * " - rootfs rootfs "
673 * IIUC, so long as we've chrooted so that rootfs is not our root,
674 * the rootfs entry should always be skipped in mountinfo contents.
676 bool detect_ramfs_rootfs(void)
684 f
= fopen("/proc/self/mountinfo", "r");
688 while (getline(&line
, &len
, f
) != -1) {
689 for (p
= line
, i
= 0; p
&& i
< 4; i
++)
690 p
= strchr(p
+ 1, ' ');
694 p2
= strchr(p
+ 1, ' ');
699 if (strcmp(p
+ 1, "/") == 0) {
700 /* This is '/'. Is it the ramfs? */
701 p
= strchr(p2
+ 1, '-');
702 if (p
&& strncmp(p
, "- rootfs rootfs ", 16) == 0) {
705 INFO("Rootfs is located on ramfs");
716 char *on_path(const char *cmd
, const char *rootfs
)
718 char *entry
= NULL
, *path
= NULL
;
719 char cmdpath
[MAXPATHLEN
];
722 path
= getenv("PATH");
730 lxc_iterate_parts (entry
, path
, ":") {
732 ret
= snprintf(cmdpath
, MAXPATHLEN
, "%s/%s/%s", rootfs
,
735 ret
= snprintf(cmdpath
, MAXPATHLEN
, "%s/%s", entry
, cmd
);
736 if (ret
< 0 || ret
>= MAXPATHLEN
)
739 if (access(cmdpath
, X_OK
) == 0) {
741 return strdup(cmdpath
);
749 bool cgns_supported(void)
751 return file_exists("/proc/self/ns/cgroup");
754 /* historically lxc-init has been under /usr/lib/lxc and under
755 * /usr/lib/$ARCH/lxc. It now lives as $prefix/sbin/init.lxc.
757 char *choose_init(const char *rootfs
)
760 const char *empty
= "",
762 int ret
, env_set
= 0;
764 if (!getenv("PATH")) {
765 if (setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 0))
766 SYSERROR("Failed to setenv");
771 retv
= on_path("init.lxc", rootfs
);
774 if (unsetenv("PATH"))
775 SYSERROR("Failed to unsetenv");
781 retv
= malloc(PATH_MAX
);
790 ret
= snprintf(retv
, PATH_MAX
, "%s/%s/%s", tmp
, SBINDIR
, "/init.lxc");
791 if (ret
< 0 || ret
>= PATH_MAX
) {
792 ERROR("pathname too long");
796 if (access(retv
, X_OK
) == 0)
799 ret
= snprintf(retv
, PATH_MAX
, "%s/%s/%s", tmp
, LXCINITDIR
, "/lxc/lxc-init");
800 if (ret
< 0 || ret
>= PATH_MAX
) {
801 ERROR("pathname too long");
805 if (access(retv
, X_OK
) == 0)
808 ret
= snprintf(retv
, PATH_MAX
, "%s/usr/lib/lxc/lxc-init", tmp
);
809 if (ret
< 0 || ret
>= PATH_MAX
) {
810 ERROR("pathname too long");
814 if (access(retv
, X_OK
) == 0)
817 ret
= snprintf(retv
, PATH_MAX
, "%s/sbin/lxc-init", tmp
);
818 if (ret
< 0 || ret
>= PATH_MAX
) {
819 ERROR("pathname too long");
823 if (access(retv
, X_OK
) == 0)
827 * Last resort, look for the statically compiled init.lxc which we
828 * hopefully bind-mounted in.
829 * If we are called during container setup, and we get to this point,
830 * then the init.lxc.static from the host will need to be bind-mounted
831 * in. So we return NULL here to indicate that.
836 ret
= snprintf(retv
, PATH_MAX
, "/init.lxc.static");
837 if (ret
< 0 || ret
>= PATH_MAX
) {
838 WARN("Nonsense - name /lxc.init.static too long");
842 if (access(retv
, X_OK
) == 0)
851 * Given the '-t' template option to lxc-create, figure out what to
852 * do. If the template is a full executable path, use that. If it
853 * is something like 'sshd', then return $templatepath/lxc-sshd.
854 * On success return the template, on error return NULL.
856 char *get_template_path(const char *t
)
861 if (t
[0] == '/' && access(t
, X_OK
) == 0) {
866 len
= strlen(LXCTEMPLATEDIR
) + strlen(t
) + strlen("/lxc-") + 1;
872 ret
= snprintf(tpath
, len
, "%s/lxc-%s", LXCTEMPLATEDIR
, t
);
873 if (ret
< 0 || ret
>= len
) {
878 if (access(tpath
, X_OK
) < 0) {
879 SYSERROR("bad template: %s", t
);
888 * @path: a pathname where / replaced with '\0'.
889 * @offsetp: pointer to int showing which path segment was last seen.
890 * Updated on return to reflect the next segment.
891 * @fulllen: full original path length.
892 * Returns a pointer to the next path segment, or NULL if done.
894 static char *get_nextpath(char *path
, int *offsetp
, int fulllen
)
896 int offset
= *offsetp
;
898 if (offset
>= fulllen
)
901 while (path
[offset
] != '\0' && offset
< fulllen
)
904 while (path
[offset
] == '\0' && offset
< fulllen
)
908 return (offset
< fulllen
) ? &path
[offset
] : NULL
;
912 * Check that @subdir is a subdir of @dir. @len is the length of
913 * @dir (to avoid having to recalculate it).
915 static bool is_subdir(const char *subdir
, const char *dir
, size_t len
)
917 size_t subdirlen
= strlen(subdir
);
922 if (strncmp(subdir
, dir
, len
) != 0)
925 if (dir
[len
-1] == '/')
928 if (subdir
[len
] == '/' || subdirlen
== len
)
935 * Check if the open fd is a symlink. Return -ELOOP if it is. Return
936 * -ENOENT if we couldn't fstat. Return 0 if the fd is ok.
938 static int check_symlink(int fd
)
943 ret
= fstat(fd
, &sb
);
947 if (S_ISLNK(sb
.st_mode
))
954 * Open a file or directory, provided that it contains no symlinks.
956 * CAVEAT: This function must not be used for other purposes than container
957 * setup before executing the container's init
959 static int open_if_safe(int dirfd
, const char *nextpath
)
961 int newfd
= openat(dirfd
, nextpath
, O_RDONLY
| O_NOFOLLOW
);
962 if (newfd
>= 0) /* Was not a symlink, all good. */
968 if (errno
== EPERM
|| errno
== EACCES
) {
969 /* We're not root (cause we got EPERM) so try opening with
972 newfd
= openat(dirfd
, nextpath
, O_PATH
| O_NOFOLLOW
);
974 /* O_PATH will return an fd for symlinks. We know
975 * nextpath wasn't a symlink at last openat, so if fd is
976 * now a link, then something * fishy is going on.
978 int ret
= check_symlink(newfd
);
990 * Open a path intending for mounting, ensuring that the final path
991 * is inside the container's rootfs.
993 * CAVEAT: This function must not be used for other purposes than container
994 * setup before executing the container's init
996 * @target: path to be opened
997 * @prefix_skip: a part of @target in which to ignore symbolic links. This
998 * would be the container's rootfs.
1000 * Return an open fd for the path, or <0 on error.
1002 static int open_without_symlink(const char *target
, const char *prefix_skip
)
1004 int curlen
= 0, dirfd
, fulllen
, i
;
1007 fulllen
= strlen(target
);
1009 /* make sure prefix-skip makes sense */
1010 if (prefix_skip
&& strlen(prefix_skip
) > 0) {
1011 curlen
= strlen(prefix_skip
);
1012 if (!is_subdir(target
, prefix_skip
, curlen
)) {
1013 ERROR("WHOA there - target '%s' didn't start with prefix '%s'",
1014 target
, prefix_skip
);
1019 * get_nextpath() expects the curlen argument to be
1020 * on a (turned into \0) / or before it, so decrement
1021 * curlen to make sure that happens
1030 /* Make a copy of target which we can hack up, and tokenize it */
1031 if ((dup
= strdup(target
)) == NULL
) {
1032 SYSERROR("Out of memory checking for symbolic link");
1036 for (i
= 0; i
< fulllen
; i
++) {
1041 dirfd
= open(prefix_skip
, O_RDONLY
);
1046 int newfd
, saved_errno
;
1049 if ((nextpath
= get_nextpath(dup
, &curlen
, fulllen
)) == NULL
)
1052 newfd
= open_if_safe(dirfd
, nextpath
);
1053 saved_errno
= errno
;
1058 errno
= saved_errno
;
1060 SYSERROR("%s in %s was a symbolic link!", nextpath
, target
);
1072 * Safely mount a path into a container, ensuring that the mount target
1073 * is under the container's @rootfs. (If @rootfs is NULL, then the container
1074 * uses the host's /)
1076 * CAVEAT: This function must not be used for other purposes than container
1077 * setup before executing the container's init
1079 int safe_mount(const char *src
, const char *dest
, const char *fstype
,
1080 unsigned long flags
, const void *data
, const char *rootfs
)
1082 int destfd
, ret
, saved_errno
;
1083 /* Only needs enough for /proc/self/fd/<fd>. */
1084 char srcbuf
[50], destbuf
[50];
1086 const char *mntsrc
= src
;
1091 /* todo - allow symlinks for relative paths if 'allowsymlinks' option is passed */
1092 if (flags
& MS_BIND
&& src
&& src
[0] != '/') {
1093 INFO("this is a relative bind mount");
1095 srcfd
= open_without_symlink(src
, NULL
);
1099 ret
= snprintf(srcbuf
, 50, "/proc/self/fd/%d", srcfd
);
1100 if (ret
< 0 || ret
> 50) {
1102 ERROR("Out of memory");
1108 destfd
= open_without_symlink(dest
, rootfs
);
1111 saved_errno
= errno
;
1113 errno
= saved_errno
;
1119 ret
= snprintf(destbuf
, 50, "/proc/self/fd/%d", destfd
);
1120 if (ret
< 0 || ret
> 50) {
1125 ERROR("Out of memory");
1129 ret
= mount(mntsrc
, destbuf
, fstype
, flags
, data
);
1130 saved_errno
= errno
;
1136 errno
= saved_errno
;
1137 SYSERROR("Failed to mount %s onto %s", src
? src
: "(null)", dest
);
1145 * Mount a proc under @rootfs if proc self points to a pid other than
1146 * my own. This is needed to have a known-good proc mount for setting
1147 * up LSMs both at container startup and attach.
1149 * @rootfs : the rootfs where proc should be mounted
1151 * Returns < 0 on failure, 0 if the correct proc was already mounted
1152 * and 1 if a new proc was mounted.
1154 * NOTE: not to be called from inside the container namespace!
1156 int lxc_mount_proc_if_needed(const char *rootfs
)
1158 char path
[MAXPATHLEN
];
1159 int link_to_pid
, linklen
, mypid
, ret
;
1160 char link
[INTTYPE_TO_STRLEN(pid_t
)] = {0};
1162 ret
= snprintf(path
, MAXPATHLEN
, "%s/proc/self", rootfs
);
1163 if (ret
< 0 || ret
>= MAXPATHLEN
) {
1164 SYSERROR("proc path name too long");
1168 linklen
= readlink(path
, link
, INTTYPE_TO_STRLEN(pid_t
));
1170 ret
= snprintf(path
, MAXPATHLEN
, "%s/proc", rootfs
);
1171 if (ret
< 0 || ret
>= MAXPATHLEN
) {
1172 SYSERROR("proc path name too long");
1176 /* /proc not mounted */
1178 if (mkdir(path
, 0755) && errno
!= EEXIST
)
1182 } else if (linklen
>= INTTYPE_TO_STRLEN(pid_t
)) {
1183 link
[linklen
- 1] = '\0';
1184 ERROR("readlink returned truncated content: \"%s\"", link
);
1188 mypid
= lxc_raw_getpid();
1189 INFO("I am %d, /proc/self points to \"%s\"", mypid
, link
);
1191 if (lxc_safe_int(link
, &link_to_pid
) < 0)
1194 /* correct procfs is already mounted */
1195 if (link_to_pid
== mypid
)
1198 ret
= umount2(path
, MNT_DETACH
);
1200 WARN("failed to umount \"%s\" with MNT_DETACH", path
);
1203 /* rootfs is NULL */
1204 if (!strcmp(rootfs
, ""))
1205 ret
= mount("proc", path
, "proc", 0, NULL
);
1207 ret
= safe_mount("proc", path
, "proc", 0, NULL
, rootfs
);
1211 INFO("mounted /proc in container for security transition");
1215 int open_devnull(void)
1217 int fd
= open("/dev/null", O_RDWR
);
1220 SYSERROR("Can't open /dev/null");
1225 int set_stdfds(int fd
)
1232 ret
= dup2(fd
, STDIN_FILENO
);
1236 ret
= dup2(fd
, STDOUT_FILENO
);
1240 ret
= dup2(fd
, STDERR_FILENO
);
1247 int null_stdfds(void)
1252 fd
= open_devnull();
1254 ret
= set_stdfds(fd
);
1261 /* Check whether a signal is blocked by a process. */
1262 /* /proc/pid-to-str/status\0 = (5 + 21 + 7 + 1) */
1263 #define __PROC_STATUS_LEN (6 + INTTYPE_TO_STRLEN(pid_t) + 7 + 1)
1264 bool task_blocks_signal(pid_t pid
, int signal
)
1267 char status
[__PROC_STATUS_LEN
];
1269 uint64_t sigblk
= 0, one
= 1;
1274 ret
= snprintf(status
, __PROC_STATUS_LEN
, "/proc/%d/status", pid
);
1275 if (ret
< 0 || ret
>= __PROC_STATUS_LEN
)
1278 f
= fopen(status
, "r");
1282 while (getline(&line
, &n
, f
) != -1) {
1285 if (strncmp(line
, "SigBlk:", 7))
1288 numstr
= lxc_trim_whitespace_in_place(line
+ 7);
1289 ret
= lxc_safe_uint64(numstr
, &sigblk
, 16);
1296 if (sigblk
& (one
<< (signal
- 1)))
1305 int lxc_preserve_ns(const int pid
, const char *ns
)
1308 /* 5 /proc + 21 /int_as_str + 3 /ns + 20 /NS_NAME + 1 \0 */
1309 #define __NS_PATH_LEN 50
1310 char path
[__NS_PATH_LEN
];
1312 /* This way we can use this function to also check whether namespaces
1313 * are supported by the kernel by passing in the NULL or the empty
1316 ret
= snprintf(path
, __NS_PATH_LEN
, "/proc/%d/ns%s%s", pid
,
1317 !ns
|| strcmp(ns
, "") == 0 ? "" : "/",
1318 !ns
|| strcmp(ns
, "") == 0 ? "" : ns
);
1319 if (ret
< 0 || (size_t)ret
>= __NS_PATH_LEN
) {
1324 return open(path
, O_RDONLY
| O_CLOEXEC
);
1327 int lxc_switch_uid_gid(uid_t uid
, gid_t gid
)
1329 if (setgid(gid
) < 0) {
1330 SYSERROR("Failed to switch to gid %d.", gid
);
1333 NOTICE("Switched to gid %d.", gid
);
1335 if (setuid(uid
) < 0) {
1336 SYSERROR("Failed to switch to uid %d.", uid
);
1339 NOTICE("Switched to uid %d.", uid
);
1344 /* Simple covenience function which enables uniform logging. */
1345 int lxc_setgroups(int size
, gid_t list
[])
1347 if (setgroups(size
, list
) < 0) {
1348 SYSERROR("Failed to setgroups().");
1351 NOTICE("Dropped additional groups.");
1356 static int lxc_get_unused_loop_dev_legacy(char *loop_name
)
1359 struct loop_info64 lo64
;
1361 int dfd
= -1, fd
= -1, ret
= -1;
1363 dir
= opendir("/dev");
1367 while ((dp
= readdir(dir
))) {
1368 if (strncmp(dp
->d_name
, "loop", 4) != 0)
1375 fd
= openat(dfd
, dp
->d_name
, O_RDWR
);
1379 ret
= ioctl(fd
, LOOP_GET_STATUS64
, &lo64
);
1381 if (ioctl(fd
, LOOP_GET_STATUS64
, &lo64
) == 0 ||
1389 ret
= snprintf(loop_name
, LO_NAME_SIZE
, "/dev/%s", dp
->d_name
);
1390 if (ret
< 0 || ret
>= LO_NAME_SIZE
) {
1407 static int lxc_get_unused_loop_dev(char *name_loop
)
1410 int fd_ctl
= -1, fd_tmp
= -1;
1412 fd_ctl
= open("/dev/loop-control", O_RDWR
| O_CLOEXEC
);
1416 loop_nr
= ioctl(fd_ctl
, LOOP_CTL_GET_FREE
);
1420 ret
= snprintf(name_loop
, LO_NAME_SIZE
, "/dev/loop%d", loop_nr
);
1421 if (ret
< 0 || ret
>= LO_NAME_SIZE
)
1424 fd_tmp
= open(name_loop
, O_RDWR
| O_CLOEXEC
);
1433 int lxc_prepare_loop_dev(const char *source
, char *loop_dev
, int flags
)
1436 struct loop_info64 lo64
;
1437 int fd_img
= -1, fret
= -1, fd_loop
= -1;
1439 fd_loop
= lxc_get_unused_loop_dev(loop_dev
);
1441 if (fd_loop
== -ENODEV
)
1442 fd_loop
= lxc_get_unused_loop_dev_legacy(loop_dev
);
1447 fd_img
= open(source
, O_RDWR
| O_CLOEXEC
);
1451 ret
= ioctl(fd_loop
, LOOP_SET_FD
, fd_img
);
1455 memset(&lo64
, 0, sizeof(lo64
));
1456 lo64
.lo_flags
= flags
;
1458 ret
= ioctl(fd_loop
, LOOP_SET_STATUS64
, &lo64
);
1468 if (fret
< 0 && fd_loop
>= 0) {
1476 int lxc_unstack_mountpoint(const char *path
, bool lazy
)
1482 ret
= umount2(path
, lazy
? MNT_DETACH
: 0);
1484 /* We consider anything else than EINVAL deadly to prevent going
1485 * into an infinite loop. (The other alternative is constantly
1486 * parsing /proc/self/mountinfo which is yucky and probably
1489 if (errno
!= EINVAL
)
1492 /* Just stop counting when this happens. That'd just be so
1493 * stupid that we won't even bother trying to report back the
1494 * correct value anymore.
1496 if (umounts
!= INT_MAX
)
1499 /* We succeeded in umounting. Make sure that there's no other
1500 * mountpoint stacked underneath.
1508 int run_command(char *buf
, size_t buf_size
, int (*child_fn
)(void *), void *args
)
1511 int ret
, fret
, pipefd
[2];
1514 /* Make sure our callers do not receive uninitialized memory. */
1515 if (buf_size
> 0 && buf
)
1518 if (pipe(pipefd
) < 0) {
1519 SYSERROR("failed to create pipe");
1523 child
= lxc_raw_clone(0);
1527 SYSERROR("failed to create new process");
1532 /* Close the read-end of the pipe. */
1535 /* Redirect std{err,out} to write-end of the
1538 ret
= dup2(pipefd
[1], STDOUT_FILENO
);
1540 ret
= dup2(pipefd
[1], STDERR_FILENO
);
1542 /* Close the write-end of the pipe. */
1546 SYSERROR("failed to duplicate std{err,out} file descriptor");
1547 _exit(EXIT_FAILURE
);
1550 /* Does not return. */
1552 ERROR("failed to exec command");
1553 _exit(EXIT_FAILURE
);
1556 /* close the write-end of the pipe */
1559 if (buf
&& buf_size
> 0) {
1560 bytes
= lxc_read_nointr(pipefd
[0], buf
, buf_size
- 1);
1562 buf
[bytes
- 1] = '\0';
1565 fret
= wait_for_pid(child
);
1566 /* close the read-end of the pipe */
1572 bool lxc_nic_exists(char *nic
)
1574 #define __LXC_SYS_CLASS_NET_LEN 15 + IFNAMSIZ + 1
1575 char path
[__LXC_SYS_CLASS_NET_LEN
];
1579 if (!strcmp(nic
, "none"))
1582 ret
= snprintf(path
, __LXC_SYS_CLASS_NET_LEN
, "/sys/class/net/%s", nic
);
1583 if (ret
< 0 || (size_t)ret
>= __LXC_SYS_CLASS_NET_LEN
)
1586 ret
= stat(path
, &sb
);
1593 uint64_t lxc_find_next_power2(uint64_t n
)
1595 /* 0 is not valid input. We return 0 to the caller since 0 is not a
1596 * valid power of two.
1611 int lxc_set_death_signal(int signal
)
1616 ret
= prctl(PR_SET_PDEATHSIG
, prctl_arg(signal
), prctl_arg(0),
1617 prctl_arg(0), prctl_arg(0));
1619 /* Check whether we have been orphaned. */
1620 ppid
= (pid_t
)syscall(SYS_getppid
);
1624 self
= lxc_raw_getpid();
1625 ret
= kill(self
, SIGKILL
);
1631 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", signal
);
1638 int fd_cloexec(int fd
, bool cloexec
)
1642 oflags
= fcntl(fd
, F_GETFD
, 0);
1647 nflags
= oflags
| FD_CLOEXEC
;
1649 nflags
= oflags
& ~FD_CLOEXEC
;
1651 if (nflags
== oflags
)
1654 if (fcntl(fd
, F_SETFD
, nflags
) < 0)
1660 int recursive_destroy(char *dirname
)
1663 struct dirent
*direntp
;
1667 dir
= opendir(dirname
);
1671 while ((direntp
= readdir(dir
))) {
1675 if (!strcmp(direntp
->d_name
, ".") ||
1676 !strcmp(direntp
->d_name
, ".."))
1679 pathname
= must_make_path(dirname
, direntp
->d_name
, NULL
);
1681 ret
= lstat(pathname
, &mystat
);
1684 WARN("Failed to stat \"%s\"", pathname
);
1690 if (!S_ISDIR(mystat
.st_mode
))
1693 ret
= recursive_destroy(pathname
);
1701 ret
= rmdir(dirname
);
1704 SYSWARN("Failed to delete \"%s\"", dirname
);
1709 ret
= closedir(dir
);
1712 SYSWARN("Failed to delete \"%s\"", dirname
);